Documentation

Mainmind is one mount for the whole company. Every agent you already pay for and every person you trust plugs into the same control plane and gets the same thing: the organization's knowledge at a named commit, the rules that bound what may happen, the live run ledger, and the judgment loop. Decisions reach the owner as one link with two buttons, and the answer comes back as standing law. Underneath, everything the mount serves is a projection of plain files in your own repository: delete the app and the company is still whole.

These pages are the public documentation, written for both kinds of reader. Agents: the index at /llms.txt lists every page in raw markdown; the machine-readable tool surface is at /api/surface.

Getting started: mount an organization from claude.ai, Claude Code, or plain curl.
Concepts: the company file, projections, runs, decisions, roles, and tenants.
MCP tools: every tool on the mount, with arguments and role gates. Generated from the same registry the server registers from.
HTTP API: the same contract over plain HTTP, for runners with no MCP support.
Auth, errors, examples: where a token comes from, what failure looks like, and a worked run from start to finish.

The pattern underneath is an open standard: Organizational Seed (Apache-2.0). The hosted product and the self-run folder are the same files.

Written by hand. The tool and endpoint references are the generated ones that cannot drift; this page is maintained like any other prose. Raw markdown for agents: /docs/index.md. Index: /llms.txt.